1. Technical Field
The present application relates generally to an improved data processing system and method. More specifically, the present application is directed to a method and procedure for detecting cable length in a storage subsystem with wide ports.
2. Description of Related Art
In storage network systems, high speed serial differential interfaces are used to interconnect multiple storage components. For example, in BladeCenter® products from IBM Corporation, a serially attached SCSI (SAS) switch may be used to interconnect the server blades to external storage, such as a typical storage enclosure. The server blades may be directly connected to the SAS switch via an internal high speed fabric. The SAS switch is connected to the external storage via external SAS cables.
Generally, multiple cable lengths are required for attaching storage at different distances from the SAS switch. The initial release of the first BladeCenter® storage product may require a “short” cable, such as three meters, and a “long” cable, such as eleven meters. Soon thereafter, the storage product may require longer cables, such as twenty meters.
As the high speed interfaces increase in data rate speed, it becomes necessary to selectively adjust the transmitter/receiver characteristics, such as pre-emphasis and de-emphasis. With significantly disparate cable lengths, it is difficult to optimize the high speed interface for both short and long cables. Therefore, it becomes necessary to determine the cable length attached to each port of the SAS switch. Furthermore, some scenarios may occur in which a short cable is inadvertently, or possibly deliberately, replaced with a long cable.
To accommodate the different cable lengths, created either by statically preplanned cabling procedures or by dynamically swapping cables in a customer location, it becomes necessary to dynamically determine cable lengths between a SAS switch and external storage. Several methods have been proposed and implemented in the prior art. For example, some fiber channel cables implement an embedded VPD (vital product data) circuit that includes cable length information. This has only been implemented with cables using small form factor pluggable (SFP) connections. Whether the cables are optical or copper, it remains that the necessary cable length information is implemented with some type of cable VPD, which is only accessible via some sort of out-of-band interface embedded within the high speed cable.
Furthermore, very recent SAS cabling technology employs the notion of a “wide” port. A wide port consists of multiple lanes or physical transceiver elements (PHYs). Today, SFPs are designed for a single port. It is quite impractical to provide a wide SFP for optical ports. For example, a four-wide port would require four laser transmitters and four receivers. Using a wide SFP for copper cabling would be more likely, but a significant cost adder would be required. It should be noted that SFPs, whether optical or copper, require an out-of-band interface, something that is heretofore not standardized or implemented.
A generally accepted bit error rate (BER) for high speed serial interfaces is 1×10−12 (one error occurrence for every 1012 bits that are transferred). Some things that can affect high speed signaling include impedance variation caused by unexpected electrical discontinuities along the transmission path, high speed driver/receiver circuit defects, improper mating contacts caused by bent or damaged connector pins, incomplete connector mating caused by mechanical or installation problems, and signal coupling between adjacent signal paths. Individual components are tested to a performance range, but tolerance buildup can cause attenuation beyond nominal design targets. Often, performance parameters are guaranteed by the manufacturing process controls and not 100% tested. Therefore, there is an exposure to maverick defects.
Ideally, all the above problems are tested and verified by interconnect and subsystem manufacturers. However, often this is not the case, and such defects are introduced into the final system integration process. All the low speed circuitry (<1 GHz) can be adequately tested. The high speed circuitry must be carefully verified. A common technique is to wrap the high speed interfaces using a cable or wrap paths external to the subsystem; however, this does not cover the actual interface connections at the time of system integration.
As subsystem components are integrated into a system, the parametric variance from the nominal case may cause a communication failure across the high speed interface. Aggravating factors may include customer data pattern, printed circuit variations and parasitics, connector parasitics, cable length or cable discontinuities, and system environment. When a communication failure is detected, the system may try to retransmit the data or may employ error correction schemes. The price for transmission recovery may be realized in lost performance. Performance degradation may be measured in bit error rate (BER).